What can a token reach?

The token decides which workspace a request works on, so no endpoint takes a workspace ID.

  • One workspace: the one it was created in.
  • Everything the API covers in that workspace: both agents’ settings and switches, knowledge and approvals, insights, full transcripts and other tokens. There are no read-only or per-channel tokens.
  • Until it’s revoked. Tokens don’t expire, and they stop working at once when they’re revoked or the owner’s account is disabled.
  • Not calls. The API has no endpoint that starts a call: visitors start them from the widget, and you run test calls from the dashboard.
  • Not the dashboard-only jobs. Phone verification for human handover, an early release, and billing work only in the signed-in dashboard, and the API refuses them when a token is used.

What do authentication errors look like?

Each failure returns 401 with a JSON body that names the problem. Over MCP, the same messages come back as a JSON-RPC error with code -32001 and the same HTTP status.

Authentication errors from the REST API
RequestStatusResponse body
No Authorization header401{"error": "Sign in to manage your workspace."}
A scheme other than Bearer401{"error": "Invalid access token."}
An unknown or revoked token401{"error": "Invalid or revoked workspace token."}

Can a web page call the API with a token?

No. The management API sends no CORS headers, so browsers block requests to it from your web pages, and a token in front-end code could be copied by anyone who opens the page. Keep tokens in server-side code, a CI secret or a trusted MCP client.

Create a separate token for each tool, so you can revoke one without breaking the others. The API key security guide covers storage, rotation and what to do after a leak.